Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

MarketGPT: Developing a Pre-trained transformer (GPT) for Modeling Financial Time Series

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A GPT trained on Nasdaq message data generates realistic stock order flow at message scale inside a discrete event simulator.

desk verdict A useful GPT-based LOB message generator proof-of-concept, but the realism claim is partly an artifact of a hand-written error-correction filter that reshapes ~7% of the output. read the letter →

arxiv 2411.16585 v1 pith:MSFG4JLZ submitted 2024-11-25 q-fin.TR

Aaron Wheeler , Jeffrey D. Varner This is my paper · ORCID
classification q-fin.TR
keywords MarketGPTgenerativepre-trainedtransformerlimitorderbookflowstylizedfactsattentionsinksdiscreteeventsimulationNasdaqITCH
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that a transformer-based generative model, trained on Nasdaq ITCH order-message data, can act as an order-generation engine inside a discrete event simulator, producing realistic limit-order-book message streams. The authors report that the generated streams reproduce key stylized facts of real markets—heavy-tailed returns, volatility clustering, long-range dependence, and realistic distributions of order sizes and inter-arrival times—at the individual-message level, and that a single attention-sink token lets the model keep generating realistic flow beyond its training context window. The intended payoff is an interactive market simulator that could be used to stress-test trading strategies or evaluate proposed market rules before they are applied.

What carries the argument

The machinery is a token-level autoregressive transformer that treats each order message as a sequence of 24 tokens drawn from a vocabulary of about 12,111 (covering symbol, side, order type, size, relative price, and timestamp fields) and is trained with cross-entropy loss to predict the next token. The model is a roughly 100-million-parameter decoder-only transformer using RMSNorm, rotary position embeddings, a KV cache, and a single dedicated attention-sink token that gives attention an explicit place to offload scores when the sliding context window moves beyond the training length. Generated messages are passed into a discrete event simulator, where an error-correction procedure checks whether referenced orders exist in the book; for the roughly 7% of messages that fail (mostly hallucinated replace orders), the simulator either substitutes the first order in the priority queue or discards the message and re-runs the timestep.

What would settle it

Re-run the same model with error correction disabled (or log raw token sequences before correction) and count: (i) the fraction of cancel/replace messages that refer to nonexistent orders, (ii) the message-type and order-size distributions of the raw stream. If a large share of referential messages are invalid or the raw distributions diverge strongly from the empirical ones, the realism is produced by the simulator's rule-based correction rather than by the learned token distribution.

Watch

Extended reading notes

Core claim

The central claim is that a GPT trained to predict the next token of a tokenized Nasdaq ITCH message stream generates 'highly realistic order flow' and, to the authors' knowledge, is the first study to validate many market stylized facts at the message scale. Acting as a conditional world agent inside a discrete event simulator, the model reproduces the distribution of message types, order sizes, and inter-arrival times; its generated mid-price returns show heavy tails (excess kurtosis about 4.9 vs. an empirical 2.4), volatility clustering, and power-law autocorrelation decay of absolute returns (DFA exponent α ≈ 0.73 vs. 0.64, Hurst exponent H ≈ 0.77 vs. 0.68), even though none of these properties appears in the loss function. The authors also claim that a dedicated sink token in the attention mechanism lets the model sustain realistic generation long after the original prompt has left the context window.

Load-bearing premise

The claim that the GPT itself generates realistic order flow rests on statistics computed after a rule-based error-correction pass that discards or rewrites roughly 7% of the model's messages; the paper does not show the raw, uncorrected output.

Editorial extensions

If this is right

  • The model can produce long synthetic streams of limit-order-book messages that satisfy market stylized facts at message scale, making it a candidate data generator for training or stress-testing trading agents without exposing real order flow.
  • Because the attention-sink streaming mechanism keeps the model fluent beyond its training context, whole trading days can be simulated from a short initial prompt rather than restarting the context.
  • Including all price levels of the book (not just the top) lets the simulator represent cancellation dynamics and order-book imbalances, which the authors argue matters for periods such as flash crashes.
  • The paper's platform is positioned as a first step toward interactive market simulators for ex-ante evaluation of market rules and for market-impact studies where a user trades against the world agent.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper does not ablate the error-correction step, so a decisive test is to compare raw, uncorrected model samples against the corrected streams; if a large fraction of raw referential messages point to nonexistent orders, the rule-based correction, not the learned token distribution, is doing the load-bearing work for realism.
  • The generated series overshoot the empirical heavy tails, volatility clustering, and spread/liquidity metrics, which suggests the model has learned price-change scaling without yet reproducing the order-book depth that gives rise to it; conditioning on a compact book state may reduce this bias.
  • The need to hand-tune temperature and top-p per finetuned ticker, and the high inference cost of 24 tokens per message, mean the framework as presented is not yet a drop-in simulator for arbitrary assets; adaptive sampling and faster tokenization (or speculative decoding) would be natural extensions.
  • The reported long-range dependence extending beyond the context window could also be probed by checking whether the persistence survives when the error-correction substitution is randomized; that would separate the model's learned dynamics from the simulator's book-keeping.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper presents MarketGPT, a roughly 100M-parameter decoder-only transformer that generates limit-order-book messages from Nasdaq ITCH data inside a discrete event simulator. The authors tokenize order messages, pretrain on multiple tickers, fine-tune per ticker, and sample autoregressively with attention sinks and a rolling KV cache. Evaluation compares generated and empirical order flow via message-type frequencies, inter-arrival times, order-size distributions, return kurtosis, volatility clustering, DFA and Hurst exponents, and price/volume trajectories. The stated central claim is that the model generates highly realistic order flow and is the first to validate many stylized facts at message scale, including beyond the training context length.

Significance. If the attribution is established, this is a useful step toward bottom-up, token-level market simulation: the model operates on full LOB message flow, includes replace orders, uses all price levels, and demonstrates stable streaming inference beyond the trained context via attention sinks. The code is publicly available under an MIT license. The empirical fidelity results are plausible and several comparisons are carefully done. However, the paper's central claim is currently weakened by the unseparated effect of a hand-written error-correction filter, by hand-tuned sampling parameters, and by acknowledged large liquidity deviations; these issues must be resolved or the conclusions substantially softened.

major comments (3)
  1. [Discrete event simulator / Results] The evaluation does not isolate the GPT's raw sampled distribution from the post-correction message stream. The DES section states that messages failing error correction are discarded (~7%) and that, when the referenced price level exists but the order time or size is invalid, 'the first order in the priority queue of that price level' is substituted. All statistics reported in Figs. 3–8 are computed after this correction. The paper's own explanation for the add-vs-replace discrepancy ('Whenever this replace order placement error occurred, the message was discarded and it is likely that a limit order was placed next') demonstrates that the filter changes the message-type distribution. Therefore the central conclusion that 'our model generates highly realistic order flow' is not established: the realism may come from the hand-written rule rather than the learned token distribution. Please report the raw sampled output statistics and/or run an ablation with and without the error-correction filter, quantifying changes in message-type frequencies, order-size and inter-arrival distributions, and return statistics.
  2. [Results / Model Limitations] The sampling temperature (1.02) and top-p (0.98) were selected by trial and error to balance fidelity and rare-event generation, and the Model Limitations section concedes that the same parameters did not work for models fine-tuned on other tickers. Because these parameters directly control tail probabilities and message-type frequencies, the claim that the stylized facts emerge without being 'explicitly part of the loss function' is overstated unless the authors provide a sensitivity analysis or a principled, parameter-free inference rule. At minimum, the dependence of each reported stylized fact on temperature and top-p should be shown.
  3. [Results, Fig. 6] The generated limit order book shows substantially different average volume at the best bid and ask levels and a lower average spread; the authors call this 'arguably the most glaring issue' and attribute other statistical deviations to it. This is not a minor caveat: it directly affects the 'highly realistic order flow' claim, since the spread and depth are core LOB properties. The conclusions should be softened unless the authors can show that this liquidity mismatch does not materially inflate the reported agreement in message-type, order-size, and return statistics.
minor comments (5)
  1. [Abstract] The phrase 'in a steaming manner' should be 'in a streaming manner'.
  2. [Fig. 8 caption] The caption contains the typo 'ans' for 'and' in 'money ans shares exchanged'.
  3. [Materials and Methods] The PyTorch reference is a placeholder ('[ ? ]'); please supply the proper citation.
  4. [Fig. 5 caption] The caption says 'Data across all simulation trials (N=100)', while the rest of the paper reports N=10 simulation trials; please clarify which is correct.
  5. [Results and Discussion] The DFA and Hurst estimates for the empirical series are given without confidence intervals or estimation details, while the generated series reports mean ± SD; please provide comparable uncertainty measures for both.

Circularity Check

1 steps flagged · score 3.0 of 10

Reported realism is partly enforced by the DES error-correction filter; the raw GPT output is not isolated, so the central attribution of realistic order flow to the model itself is partially circular.

  1. fitted input called prediction [Discrete event simulator; Simulated order flow (Fig. 3a discussion)]
    "If a message did not pass the error correction procedure, then the message was discarded and the simulator would re-run the timestep. This would occur for a relatively low number of messages (~7%). ... Whenever this replace order placement error occurred, the message was discarded and it is likely that a limit order was placed next (since it is the most abundant message type in the training data)—so this may explain why limit orders are placed more frequently at roughly the same frequency that replace orders are placed less."

    All reported evaluation statistics (message-type frequencies, order-size and inter-arrival-time distributions, return kurtosis, volatility clustering, long-range dependence) are computed on the post-correction message stream, not on the transformer's raw sampled output. The error-correction procedure is a hand-coded input that discards or rewrites invalid referential messages, and the authors' own explanation shows it increases add frequencies and decreases replace frequencies. Thus the 'realistic order flow' attributed to the model is in part produced by this fixed rule, not by the learned token distribution.

full rationale

The paper does not contain a formal derivation chain with equations that could be circular by construction; no load-bearing self-citation chain, imported uniqueness theorem, or renamed known result is present. The only meaningful circularity concern is empirical: the GPT is embedded in a discrete event simulator whose error-correction step modifies roughly 7% of messages, and all validation statistics are computed after this correction. The authors explicitly link the add-versus-replace frequency discrepancy to this filter, showing that the filter changes the very message-type distribution used as evidence of model realism. Additionally, the sampling temperature (1.02) and top-p (0.98) were hand-tuned by trial and error to balance fidelity and rare-event generation, so some of the observed match to the target distribution is selected for rather than independently predicted. However, the heavy-tailed returns, volatility clustering, and long-range dependence emerge from the simulated book dynamics rather than being directly hard-coded by the correction rule, so the model still contributes independent content. The finding is a partial attribution gap, not a full by-construction equivalence, hence a moderate score of 3.

Assumptions & free parameters 7 free parameters · 6 assumptions · 0 invented entities

The central claim rests on the fidelity of the ITCH data reconstruction, on the tokenization scheme preserving market statistics, on the transformer's ability to learn the message distribution, on the validity of the selected stylized facts as realism measures, and on the assumption that attention sinks enable valid long-context generation. The model also depends on hand-set preprocessing truncations and sampling parameters listed above; these are not derived from first principles.

free parameters (7)
  • temperature (sampling) = 1.02
    Hand-tuned by trial and error to balance fidelity and rare event generation; did not generalize to other tickers, so it is a fitted constant for the AAPL evaluation.
  • top-p (nucleus sampling) = 0.98
    Chosen jointly with temperature; restricts sampling to the top cumulative probability tokens and shapes the tail of the generated distribution.
  • inference context length = L = 2,688 messages
    Reduced from the training context (L = 10,368 tokens, 432 messages) to lower inference time; authors claim no substantial performance drop but do not report a systematic comparison.
  • relative price truncation = 999 ticks from mid-price
    Rare price ticks beyond 999 from the mid-price are truncated to maintain a finite vocabulary, capping the relative price tail.
  • order size truncation = 9,999 shares
    Order sizes above 9,999 are truncated to maintain a finite vocabulary, capping the order-size tail.
  • error correction filter = approximately 7% of messages discarded/rewritten
    A hand-designed rule that removes or rewrites generated messages referring to nonexistent orders; it materially shapes the output distribution before evaluation, so it functions as an ad hoc constraint.
  • architecture hyperparameters (12 layers, 12 heads, 768 embedding dims) = 12/12/768
    Selected via an unreported random grid search; the search ranges and other hyperparameters (learning rate, batch size, epochs) are not listed.
assumptions (6)
  • domain assumption The Nasdaq TotalView-ITCH 5.0 message feed, after preprocessing, contains all information needed to reconstruct the limit order book faithfully enough for evaluating stylized facts.
    The paper relies on the ITCH feed and its own reconstruction of the book; errors in parsing or in the excluded fields (hidden orders, auctions) would bias the evaluations.
  • domain assumption Tokenizing message fields into discrete vocabularies (price as ticks from mid-price, inter-arrival times, sizes) preserves the statistical properties of order flow.
    The authors adopt and extend the tokenization of Nagy et al.; the truncation of tails and discretization could remove the very heavy tails the paper claims to reproduce.
  • domain assumption An autoregressive transformer trained with next-token cross-entropy can learn the conditional message distribution sufficiently to generate coherent order flow.
    The entire method rests on this ML hypothesis; no proof or comparison to other architectures on the same data is given.
  • domain assumption The stylized facts used (heavy tails, volatility clustering, long-range dependence) are sufficient and appropriate measures of 'realistic order flow'.
    The paper does not test other important properties (order lifetimes, intraday seasonality, cross-asset correlations) and admits these are left for future work.
  • domain assumption Attention sinks plus a rolling KV cache allow valid generation beyond the training context length.
    The authors adopt attention sinks from prior work; they do not ablate the sink token to show it is responsible for the long-range fidelity.
  • standard math Standard probabilistic and optimization background (softmax, cross-entropy, Adam) is valid.
    These are used implicitly in training and sampling.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MarketGPT: Developing a Pre-trained transformer (GPT) for Modeling Financial Time Series." pith.science (2026). https://pith.science/paper/MSFG4JLZ

@misc{pith2026241116585,
  author       = {Pith},
  title        = {Pith review of: MarketGPT: Developing a Pre-trained transformer (GPT) for Modeling Financial Time Series},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MSFG4JLZ}},
  note         = {Machine review of arXiv:2411.16585}
}
read the original abstract

This work presents a generative pre-trained transformer (GPT) designed for modeling financial time series. The GPT functions as an order generation engine within a discrete event simulator, enabling realistic replication of limit order book dynamics. Our model leverages recent advancements in large language models to produce long sequences of order messages in a steaming manner. Our results demonstrate that the model successfully reproduces key features of order flow data, even when the initial order flow prompt is no longer present within the model's context window. Moreover, evaluations reveal that the model captures several statistical properties, or 'stylized facts', characteristic of real financial markets and broader macro-scale data distributions. Collectively, this work marks a significant step toward creating high-fidelity, interactive market simulations.

Figures

Figures reproduced from arXiv: 2411.16585 by the authors.

Figure 1
Figure 1. Tokenization scheme that translates pre-processed messages (top) into encoded messages (bottom). The vocabulary fields and corresponding valid token values are listed below the diagram. Pre-processed message fields that are not tokenized (e.g., order ID and absolute price) are excluded from the figure. Fields that are similar (e.g., ∆t and time) share vocabulary values. The first three tokens are reserved for specia… view at source ↗
Figure 2
Figure 2. Schematic of the simulation platform. The token generation loop contains the prompt (previous messages up to context length) and the world agent—i.e., the transformer model. Upon sampling the encoded message length (24 tokens), the error correction procedure checks for and remedies hallucination conditions if possible. If the generated message is deemed valid by the error correction procedure, the discrete event sim… view at source ↗
Figure 3
Figure 3. (a) Generated and empirical message type frequencies. Data across all simulation trials (N=10) were compared against the test distribution—error bars account for 95% confidence intervals. The frequencies roughly match, with the only discrepancies occurring for add and replace message types. (b) Distribution of 1-second returns of both generated and empirical messages. Data across all simulation trials (N=10) were co… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: The generated and empirical distributions of inter-arrival times on a semi-log scale (each histogram was composed of 500 bins). Data across all simulation trials (N=10) were similar and compared against the test distribution. Overlap between distributions (brownish col…
Figure 5
Figure 5. Figure 5: The generated and empirical distributions of order sizes on a log scale (each histogram was composed of 500 bins). Data across all simulation trials (N=100) were similar and compared against the test distri￾bution. Overlap between distributions (brownish color) is indi…
Figure 6
Figure 6. Figure 6: Spreads and volume offered at best bid and ask sides were averaged across 1-second intervals. (a) The average volume offered at the best bid price level. (b) The average volume offered at the best ask price level. (c) The spread, or difference between the best bid and …
Figure 7
Figure 7. Figure 7: Volatility clustering of generated and empirical returns. Data from a single trial is illustrated for clarity but the property was nearly identical across all simulation trials (N=10). (a) The autocorrelation of squared returns. (b) The autocorrelation of the absolute …
Figure 8
Figure 8. Figure 8: Prediction of (a) cumulative dollars and (b) shares traded for generated (grey) and empirical (red) time series data. The notional values of money ans shares exchanged were accurately predicted for most simulation trials. Each sequence is truncated by the minimum numbe…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Prospects of Imitating Trading Agents in the Stock Market

    q-fin.CP 2025-08 conditional novelty 4.0 of 10

    A state-space generative model trained on synthetic limit order book data can partially imitate trading agent behavior, matching some action distributions while underestimating cancellations, with results limited by h...

Reference graph

Works this paper leans on

57 extracted references · 34 canonical work pages · cited by 1 Pith paper

  1. [1]

    Journal of Financial Economics 7: 229–263

    Cox JC, Ross SA, Rubinstein M (1979) Option pricing: A simplified approach. Journal of Financial Economics 7: 229–263

    work page 1979

  2. [2]

    Journal of Economic Perspectives 15: 157–168

    Engle R (2001) GARCH 101: The use of ARCH/GARCH models in applied economet- rics. Journal of Economic Perspectives 15: 157–168

    work page 2001

  3. [3]

    In: Time Series and Statistics, Palgrave Macmil- lan UK

    Harvey AC (1990) ARIMA models. In: Time Series and Statistics, Palgrave Macmil- lan UK. pp. 22–24. doi:10.1007/978-1-349-20865-4 2. URL https://doi.org/10.1007/ 978-1-349-20865-4 2

    work page doi:10.1007/978-1-349-20865-4 1990

  4. [4]

    Journal of Economic Perspectives 15: 101–115

    Stock JH, Watson MW (2001) Vector autoregressions. Journal of Economic Perspectives 15: 101–115

    work page 2001

  5. [5]

    Journal of Political Economy 81: 637–654

    Black F, Scholes M (1973) The pricing of options and corporate liabilities. Journal of Political Economy 81: 637–654

    work page 1973

  6. [6]

    (2014) Gener- ative adversarial nets

    Goodfellow IJ, Pouget-Abadie J, Mirza M, Xu B, Warde-Farley D, et al. (2014) Gener- ative adversarial nets. In: Proceedings of the 27th International Conference on Neural Information Processing Systems - Volume 2. Cambridge, MA, USA: MIT Press, NIPS’14, pp. 2672–2680

    work page 2014

  7. [7]

    Generative adversarial networks in finance: an overview

    Eckerli F, Osterrieder J (2021). Generative adversarial networks in finance: an overview. doi:10.48550/ARXIV.2106.06364. URL https://arxiv.org/abs/2106.06364

    arXiv doi:10.48550/arxiv.2106.06364 2021

  8. [8]

    In: Proceedings of the 2020 ACM SIGSIM Conference on Principles of Advanced Discrete Simulation

    Byrd D, Hybinette M, Balch TH (2020) Abides: Towards high-fidelity multi-agent mar- ket simulation. In: Proceedings of the 2020 ACM SIGSIM Conference on Principles of Advanced Discrete Simulation. ACM, SIGSIM-PADS ’20. doi:10.1145/3384441.3395986. URL http://dx.doi.org/10.1145/3384441.3395986

    arXiv 2020

Show all 57 references
  1. [9]

    Computational Economics

    Lussange J, Vrizzi S, Bourgeois-Gironde S, Palminteri S, Gutkin B (2022) Stock price formation: Precepts from a multi-agent reinforcement learning model. Computational Economics

    2022

  2. [10]

    Simulation and estimation of an agent-based market-model with a matching engine

    Jericevich I, Chang P, Gebbie T (2021). Simulation and estimation of an agent-based market-model with a matching engine. doi:10.48550/ARXIV.2108.07806. URL https: //arxiv.org/abs/2108.07806

    2021 doi arXiv

  3. [11]

    Scalable agent-based modeling for complex financial market simulations

    Wheeler A, Varner JD (2023). Scalable agent-based modeling for complex financial market simulations. doi:10.48550/ARXIV.2312.14903. URL https://arxiv.org/abs/2312. 14903

    2023 doi arXiv

  4. [12]

    Limit order book simulations: A review

    Jain K, Firoozye N, Kochems J, Treleaven P (2024). Limit order book simulations: A review. doi:10.48550/ARXIV.2402.17359. URL https://arxiv.org/abs/2402.17359

    2024 doi arXiv

  5. [13]

    Quantitative Finance 23: 931–958

    Hultin H, Hult H, Proutiere A, Samama S, Tarighati A (2023) A generative model of a limit order book using recurrent neural networks. Quantitative Finance 23: 931–958

    2023

  6. [15]

    In: Proceedings of the 14th conference on Computational linguistics -

    Webster JJ, Kit C (1992) Tokenization as the initial phase in nlp. In: Proceedings of the 14th conference on Computational linguistics -. Association for Computational Lin- 14 guistics. doi:10.3115/992424.992434. URL http://dx.doi.org/10.3115/992424.992434

    1992

  7. [16]

    (2017) Attention is all you need

    Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, et al. (2017) Attention is all you need. In: Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook, NY, USA: Curran Associates Inc., NIPS’17, pp. 6000– 6010

    2017

  8. [17]

    Quantitative Finance 1: 223–236

    Cont R (2001) Empirical properties of asset returns: stylized facts and statistical issues. Quantitative Finance 1: 223–236

    2001

  9. [18]

    empir- ical facts

    Chakraborti A, Toke IM, Patriarca M, Abergel F (2011) Econophysics review: I. empir- ical facts. Quantitative Finance 11: 991–1012

    2011

  10. [19]

    (2013) Limit order books

    Gould MD, Porter MA, Williams S, McDonald M, Fenn DJ, et al. (2013) Limit order books. Quantitative Finance 13: 1709–1742

    2013

  11. [20]

    (2020) Get real: realism metrics for robust limit order book market simulations

    Vyetrenko S, Byrd D, Petosa N, Mahfouz M, Dervovic D, et al. (2020) Get real: realism metrics for robust limit order book market simulations. In: Proceedings of the First ACM International Conference on AI in Finance. ACM, ICAIF ’20. doi:10.1145/3383455. 3422561. URL http://dx...

    2020

  12. [21]

    martinobdl/itch: Itch50converter

    Bernasconi-De-Luca M, Fusco L, Dragi´ c O (2021). martinobdl/itch: Itch50converter. doi:10.5281/ZENODO.5209267. URL https://zenodo.org/record/5209267

    2021

  13. [23]

    In: 4th ACM International Conference on AI in Finance

    Coletta A, Jerome J, Savani R, Vyetrenko S (2023) Conditional generators for limit order book environments: Explainability, challenges, and robustness. In: 4th ACM International Conference on AI in Finance. ACM, ICAIF ’23. doi:10.1145/3604237. 3626854. URL http://dx.doi.org/10...

    2023

  14. [24]

    SSRN Electronic Journal

    Hautsch N, Huang R (2011) Limit order flow, market impact and optimal order sizes: Evidence from nasdaq totalview-itch data. SSRN Electronic Journal

    2011

  15. [25]

    SSRN Electronic Journal

    Huang R, Polak T (2011) Lobster: Limit order book reconstruction system. SSRN Electronic Journal

    2011

  16. [26]

    Zhao WX, Zhou K, Li J, Tang T, Wang X, et al. (2023). A survey of large language models. doi:10.48550/ARXIV.2303.18223. URL https://arxiv.org/abs/2303.18223

    2023 doi arXiv

  17. [27]

    Adam: A method for stochastic optimization

    Kingma DP, Ba J (2014). Adam: A method for stochastic optimization. doi:10.48550/ ARXIV.1412.6980. URL https://arxiv.org/abs/1412.6980

    2014 doi arXiv

  18. [28]

    Flashattention: Fast and memory- efficient exact attention with io-awareness

    Dao T, Fu DY, Ermon S, Rudra A, R´ e C (2022). Flashattention: Fast and memory- efficient exact attention with io-awareness. doi:10.48550/ARXIV.2205.14135. URL https: //arxiv.org/abs/2205.14135

    2022 doi arXiv

  19. [29]

    J Mach Learn Res 13: 281–305

    Bergstra J, Bengio Y (2012) Random search for hyper-parameter optimization. J Mach Learn Res 13: 281–305

    2012

  20. [30]

    In: Ad- vances in Neural Information Processing Systems 32

    Zhang B, Sennrich R (2019) Root Mean Square Layer Normalization. In: Ad- vances in Neural Information Processing Systems 32. Vancouver, Canada. URL https: //openreview.net/references/pdf?id=S1qBAf6rr

    2019

  21. [31]

    Su J, Lu Y, Pan S, Murtadha A, Wen B, et al. (2021). Roformer: Enhanced transformer 15 with rotary position embedding. doi:10.48550/ARXIV.2104.09864. URL https://arxiv. org/abs/2104.09864

    2021 doi arXiv

  22. [32]

    Ainslie J, Lee-Thorp J, de Jong M, Zemlyanskiy Y, Lebr´ on F, et al. (2023). Gqa: Training generalized multi-query transformer models from multi-head checkpoints. doi: 10.48550/ARXIV.2305.13245. URL https://arxiv.org/abs/2305.13245

    2023 doi arXiv

  23. [33]

    Pope R, Douglas S, Chowdhery A, Devlin J, Bradbury J, et al. (2022). Efficiently scaling transformer inference. doi:10.48550/ARXIV.2211.05102. URL https://arxiv.org/abs/ 2211.05102

    2022 doi arXiv

  24. [34]

    In: International Conference on Learning Represen- tations

    Press O, Smith N, Lewis M (2022) Train short, test long: Attention with linear biases enables input length extrapolation. In: International Conference on Learning Represen- tations. URL https://openreview.net/forum?id=R8sQPpGCv0

    2022

  25. [35]

    Extending context window of large language models via positional interpolation

    Chen S, Wong S, Chen L, Tian Y (2023). Extending context window of large language models via positional interpolation. doi:10.48550/ARXIV.2306.15595. URL https:// arxiv.org/abs/2306.15595

    2023 doi arXiv

  26. [36]

    Longformer: The long-document transformer

    Beltagy I, Peters ME, Cohan A (2020). Longformer: The long-document transformer. doi:10.48550/ARXIV.2004.05150. URL https://arxiv.org/abs/2004.05150

    2020 doi arXiv

  27. [37]

    Efficient streaming language models with attention sinks

    Xiao G, Tian Y, Chen B, Han S, Lewis M (2023). Efficient streaming language models with attention sinks. doi:10.48550/ARXIV.2309.17453. URL https://arxiv.org/abs/ 2309.17453

    2023 doi arXiv

  28. [38]

    In: Gurevych I, Miyao Y, editors, Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)

    Fan A, Lewis M, Dauphin Y (2018) Hierarchical neural story generation. In: Gurevych I, Miyao Y, editors, Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Melbourne, Australia: Association for Computational Linguis...

    2018 doi

  29. [39]

    In: International Conference on Learning Representations

    Holtzman A, Buys J, Du L, Forbes M, Choi Y (2020) The curious case of neural text degeneration. In: International Conference on Learning Representations. URL https: //openreview.net/forum?id=rygGQyrFvH

    2020

  30. [40]

    (2021) Abides-gym: gym environments for multi-agent discrete event simulation and application to financial markets

    Amrouni S, Moulin A, Vann J, Vyetrenko S, Balch T, et al. (2021) Abides-gym: gym environments for multi-agent discrete event simulation and application to financial markets. In: Proceedings of the Second ACM International Conference on AI in Fi- nance. ACM, ICAIF’21. doi:10.11...

    2021

  31. [41]

    Physica A: Statistical Mechanics and its Applications 300: 285–299

    Challet D, Stinchcombe R (2001) Analyzing and modeling 1+1d markets. Physica A: Statistical Mechanics and its Applications 300: 285–299

    2001

  32. [42]

    The Journal of Business 36: 394

    Mandelbrot B (1963) The variation of certain speculative prices. The Journal of Business 36: 394

    1963

  33. [43]

    In: Fractals in Engineering, Springer-Verlag

    Cont R Long range dependence in financial markets. In: Fractals in Engineering, Springer-Verlag. pp. 159–179. doi:10.1007/1-84628-048-6 11. URL https://doi.org/10. 1007/1-84628-048-6 11

    doi

  34. [44]

    (1994) Mosaic organi- zation of dna nucleotides

    Peng CK, Buldyrev SV, Havlin S, Simons M, Stanley HE, et al. (1994) Mosaic organi- zation of dna nucleotides. Physical Review E 49: 1685–1689

    1994

  35. [45]

    (2012) Detrended 16 fluctuation analysis: A scale-free view on neuronal oscillations

    Hardstone R, Poil SS, Schiavone G, Jansen R, Nikulin VV, et al. (2012) Detrended 16 fluctuation analysis: A scale-free view on neuronal oscillations. Frontiers in Physiology 3

    2012

  36. [46]

    (1995) Long- range correlation properties of coding and noncoding dna sequences: Genbank analysis

    Buldyrev SV, Goldberger AL, Havlin S, Mantegna RN, Matsa ME, et al. (1995) Long- range correlation properties of coding and noncoding dna sequences: Genbank analysis. Physical Review E 51: 5084–5091

    1995

  37. [47]

    SSRN Electronic Journal

    Cont R, Potters M, Bouchaud JP (1997) Scaling in stock market data: Stable laws and beyond. SSRN Electronic Journal

    1997

  38. [48]

    Physica A: Statistical Mechanics and its Applications 245: 437–440

    Liu Y, Cizeau P, Meyer M, Peng CK, Eugene Stanley H (1997) Correlations in economic time series. Physica A: Statistical Mechanics and its Applications 245: 437–440

    1997

  39. [49]

    Physical Review E 60: 5305– 5316

    Gopikrishnan P, Plerou V, Amaral LAN, Meyer M, Stanley HE (1999) Scaling of the distribution of fluctuations of financial market indices. Physical Review E 60: 5305– 5316

    1999

  40. [50]

    Transactions of the American Society of Civil Engineers 116: 770–799

    Hurst HE (1951) Long-term storage capacity of reservoirs. Transactions of the American Society of Civil Engineers 116: 770–799

    1951

  41. [51]

    Biometrika 63: 111–116

    ANNIS AA, LLOYD EH (1976) The expected value of the adjusted rescaled hurst range of independent normal summands. Biometrika 63: 111–116

    1976

  42. [52]

    In: Econometrica

    Mandelbrot B (1971) Analysis of long-run dependence in economics-r/s technique. In: Econometrica. WILEY-BLACKWELL 111 RIVER ST, HOBOKEN 07030-5774, NJ USA, volume 39, pp. 68–+

    1971

  43. [53]

    Gholami A, Kim S, Dong Z, Yao Z, Mahoney MW, et al. (2021). A survey of quantization methods for efficient neural network inference. doi:10.48550/ARXIV.2103.13630. URL https://arxiv.org/abs/2103.13630

    2021 doi arXiv

  44. [54]

    Fast inference from transformers via spec- ulative decoding

    Leviathan Y, Kalman M, Matias Y (2022). Fast inference from transformers via spec- ulative decoding. doi:10.48550/ARXIV.2211.17192. URL https://arxiv.org/abs/2211. 17192

    2022 doi arXiv

  45. [55]

    Hu EJ, Shen Y, Wallis P, Allen-Zhu Z, Li Y, et al. (2021). Lora: Low-rank adaptation of large language models. doi:10.48550/ARXIV.2106.09685. URL https://arxiv.org/abs/ 2106.09685

    2021 doi arXiv

  46. [56]

    Qlora: Efficient finetuning of quantized llms

    Dettmers T, Pagnoni A, Holtzman A, Zettlemoyer L (2023). Qlora: Efficient finetuning of quantized llms. doi:10.48550/ARXIV.2305.14314. URL https://arxiv.org/abs/2305. 14314

    2023 doi arXiv

  47. [57]

    Mamba: Linear-time sequence modeling with selective state spaces

    Gu A, Dao T (2023). Mamba: Linear-time sequence modeling with selective state spaces. doi:10.48550/ARXIV.2312.00752. URL https://arxiv.org/abs/2312.00752

    2023 doi arXiv

  48. [58]

    Lieber O, Lenz B, Bata H, Cohen G, Osin J, et al. (2024). Jamba: A hybrid transformer- mamba language model. doi:10.48550/ARXIV.2403.19887. URL https://arxiv.org/abs/ 2403.19887

    2024 doi arXiv

  49. [59]

    Ren L, Liu Y, Lu Y, Shen Y, Liang C, et al. (2024). Samba: Simple hybrid state space models for efficient unlimited context language modeling. doi:10.48550/ARXIV.2406. 07522. URL https://arxiv.org/abs/2406.07522. 17 1201810061201012008110101108100833207140391557237198141200811...

    2024 doi arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.